Ambient Conditioning

What Are They Telling the Children?

I was watching the Olympics with my wife the other night. Our eight-year-old was already in bed. We had subscribed to the Peacock streaming service for the sole purpose of watching the Olympic sporting extravaganza. We subscribed well prior to the anti-Christian opening ceremonies debacle. I’m not going to unpack all that was wrong with the open ceremonies, but I will observe that there is a purpose behind exposing us, not only to the explicitly offensive message, but to the ugliness inherent in the imagery. C.S. Lewis observed, in his prescient novel, That Hideous Strength, that there is malevolence involved when someone presents the patina of beauty, but withholds actual beauty upon closer inspection. “The peculiar ugliness consisted in the very fact that they kept on suggesting it and then frustrating the expectation thus aroused.” In the novel, the protagonist is being forced to occupy a room whose construction is slightly off kilter. Arches over doorways that don’t quite line up. Odd proportions between room size and ceiling height. The walls are covered in pictures which, on close inspection, are hideous and bizarre. The protagonist notices the pictures. He is “surprised at the predominance of scriptural themes. It was only at the second or third glance that one discovered certain unaccountable details - something odd about the positions of the figures’ feet or the arrangement of their fingers or the grouping. And who was the person standing between the Christ and Lazarus? And why were there so many beetles under the table at the Last Supper?” If you were already familiar with Lewis’ novel, you might have found the Olympic opening ceremonies to be a kind of deja vu experience.

As we watched the Olympics a few days after the opening ceremonies, we were subjected at one point to a commercial from, if I recall correctly, a healthcare company. Notable about the commercial was a prevalent sequence of romantically attached same-sex couples. It also contained, as the voiceover waxed eloquent about the virtues of the company, video clips of what seemed at first glance to be women - although the viewer could not help noticing something peculiar about these “women”: they projected a distinctly masculine vibe. They were abnormally tall and angular, as if Lurch, from The Addams Family, had taken up doing commercials while dressed in drag. At some point the light dawned that, in addition to being subjected to images of same-sex couples, we were being conditioned by transgender imagery.

I recounted, in an earlier post, how at least one of the later Star Wars installments subjected its viewers to imagery of two women passionately kissing each other. In a similar vein, the commercial we were watching during the Olympics concluded with the image of two bearded men kissing each other on the lips.

The salient point I want to make here is that there is a very intentional effort being made to condition the consumers of media in certain directions regarding human sexuality. I make no claims of conspiracy in this regard. No conspiracy is required when there is a general worldview consensus already in place among media professionals. Much of this conditioning isn’t propositional or verbally explicit. It often consists of merely exposing the viewer to fleeting imagery as a means of working on, and attenuating, the viewer’s subjective unease. The goal is to recalibrate our conception of what is normal.

Here again, the protagonist in That Hideous Strength offers some prescient insight:

As the desert first teaches men to love water, or as absence first reveals affection, there rose up against this background of the sour and the crooked some kind of vision of the sweet and the straight. Something else — something he vaguely called the “Normal” — apparently existed. He had never thought about it before. But there it was - solid, massive, with a shape of its own, almost like something you could touch, or eat, or fall in love with…He was not thinking in moral terms at all, or else (what is much the same thing) he was having his first deeply moral experience. He was choosing a side: the Normal.

Prior to my recent experience with the Olympic games, I had been puzzling over how it came to be that so many young people seem to blithely accept the fungibility of human sex.

Mary Harrington

has speculated that this might be an artifact of young people having grown up online, where the identity they project can be completely fluid and changeable. Their digital avatars, she has suggested, help them to conceive of their respective bodies as a sort of “meat avatar”. This is a keen observation on Harrington’s part and well worth reflecting on. But my own immediate concern, as someone who supervises the entertainment choices of an 8-year-old who is not “online”, has been to have a better grasp of if, or how, children’s TV is being used to condition children in certain directions regarding human sexuality and gender. I wanted to get beyond simply promoting outrage at anecdotal TV episodes, and try to develop a more comprehensive sense of what children are being told, and how they are being told. I have written before about my perceptions regarding the general emasculation of the subject matter in children’s TV. Though it is unhelpful to empty plot lines of masculine aspirations, the widespread delusional ideas held by youth, where sex is concerned, seem to have progressed far beyond the more narrow problem of sissified cartoons.

One of the challenges we face, in getting our arms around what is being said and done in children’s entertainment, is the sheer volume of content. How can you possibly characterize the state of play without, on the one hand, devoting your life to analyzing content or, alternatively, hiring rooms full of analysts to watch tv shows or read scripts? Manually undertaking such an effort is necessarily infeasible in terms of both time and money. At least, I myself do not have an unlimited budget. So there’s that.

Methodology

Recent advances in artificial intelligence (AI) potentially offer a way around the challenge of characterizing children’s programming at a large enough scale to develop a marginally comprehensive analysis of what is really going on. I have been experimenting with the possibilities and will describe what I have found. What follows in this section is a technical-ish description of how I went about this work - for anyone who is interested in that sort of thing. For the reader who is uninterested, he should skip ahead to the Results section below, where I unpack what I found and reflect on what I have learned.

To facilitate the analysis I want, I have been developing a software tool which allows me to automatically hoover up targeted children’s TV content and leverage public facing AI language models for performing automated analysis and summarization of the content of these shows. The content that I’m looking for originates in a variety of forms, but I want the language model(s) to ultimately process the underlying textual transcript. The dialog is what I’m addressing in this particular effort. That means this work really only partially addresses the possible sources of conditioning in entertainment. The examples I gave at the start of this post all took the form of visual expressions. But nothing about the work I’m describing here addresses the challenge of doing image analysis on, say, video streams. That can certainly be done, but it is far more expensive than processing text, and I had to limit my financial exposure. As it is, merely processing the text at a modest scale has involved almost $1000 in out-of-pocket computational costs.

The tool(s) I have developed can transcribe audio tracks from videos if closed-caption information is unavailable. But they attempt to use any pre-existing closed-caption data because that avoids the computational overhead, and cost, of audio transcription.

During the course of this experiment, I have accumulated over 11,000 scripts from a variety of children’s television shows.

• Blue’s Clues - 593 episodes

• My Little Pony - 2924 episodes

• SpongeBob SquarePants - 1305 episodes

• Bluey - 677 episodes

• Peppa Pig - 3306 episodes

• Sofia the First - 62 episodes

• Disney Jr. Various - 2931

I decided to initially focus on getting the AI model(s) to summarize the scripts and characterize whether, and how, the scripts addressed questions of human sexuality, gender, and family structure. Also, I wanted the AI to report if the script contained any blatant anti-Christian stereotypes. This is the language model prompt I eventually settled on for doing bulk characterization:

"Analyze the following text and respond with a JSON object containing three keys: summary': a brief summary of the main subject of the text, 'flag': a 'yes' or 'no' answer indicating whether the text contains any of the following subjects: changing gender, a character with more than one mother, a character with more than one father, homosexuality, anti-christian stereotypes, masculine girls, or feminine boys, 'context:' the reason for the value you assigned the 'flag' field. Text: {script text here}"

I have barely scratched the surface on the question of prompt engineering. There is a lot of experimentation and refinement to be done here. Also, there can be a diminishing return when increasing prompt complexity, as far as the results the model produces are concerned. In my case, I was trying to do as much as I possibly could in a single prompt in order to budget the number of requests I had to make to the AI model. Each request costs money. I suspect that, to achieve optimal results, one would need to make multiple passes at the data to winnow down false positives, and bucket the results more cleanly. There are ways to optimize costs by varying your approach to processing the data.

When I first started, I was spending money at an alarming rate as I experimented with various prompts. I eventually ended up spending a non-trivial amount of time putting features into the tools that let me pre-estimate costs for the analyses I intended to run. The screencast video below shows this cost estimation at work. The cost to do a single evaluation of my accumulated episodes of Peppa Pig is over $300.

Using the prompt I included above, various language models will produce something along the lines of the following two examples.

{"summary": "The text is a collection of various scenes from the children's show Peppa Pig, depicting different adventures and activities involving Peppa, George, and their family and friends. The selected scenes include playing musical instruments, a visit to the vet, a bus ride, making pottery, and many more everyday activities presented in a whimsical and educational manner suited for young children.", "flag": "no", "context": "The text is a transcript of several episodes of Peppa Pig, a children's show. It does not contain any of the specified subjects such as changing gender, characters with more than one mother or father, homosexuality, anti-Christian stereotypes, masculine girls, or feminine boys.", "id": ""}

{"summary": "The text describes various adventures and activities of Peppa Pig and her friends and family. These include attending different themed parties, engaging in outdoor activities like climbing, visits to unique destinations like Hollywood, and participating in fun and educational experiences at their playgroup. Each episode involves imaginative play, family interactions, and learning moments, often with humorous outcomes.", "flag": "yes", "context": "The text mentions a character, Penny Polar Bear, who lives with 'my mommy and my other mommy,' indicating that she has more than one mother.", "id": ""}

For the bulk results generated by AI, I have done some random auditing to compare what the AI produced with my own evaluation of selected scripts. I have found generally good results, although they are not perfect. There are random false positives (i.e. episodes flagged in error) and one suspects that means there are corresponding false negatives (i.e. episodes that should have been which were not.) But there are also - and this has given me some food for thought - episodes flagged by the AI that I myself would not have flagged at first glance. But, after reading the flagged episodes and giving it some thought, I came to the conclusion that there were clearly gender-blurring subtleties that were worth flagging for human consideration. I unpack this issue a little bit more later on.

By-and-large, the results seem to be correct enough that one might tentatively draw quantitative conclusions. Some basic experimentation reveals that refinement to my own prompts can reduce the occurrence of false positives. More qualitative conclusions certainly require real human observation and discernment.

Results

The chart below shows the percentage of episodes flagged, by the AI, of the total episodes I have collected for each TV series.

Before those readers who love Bluey completely despair, I want to say there are widely varying degrees of egregiousness in the episodes which were flagged. Some episodes were flagged merely because they dealt, directly or indirectly, with the subject matter. But an episode being flagged doesn’t necessarily mean the episode is problematic. One must read the reason provided by the AI to get a sense for why it made the cut.

What we can learn from the chart, however, is that that all of these children’s shows maintain a drip-drip of episodes that implicate family structure, gender, or human sexuality. At one level, we might be encouraged by the fact that this is not the predominate theme in any of the series I evaluated. However, I’m mindful of the fact that periodic indoctrination is harder to track and catch. Not many parents have time to watch all of their children’s programs with them. So any parent who watches these shows would, on average, have to watch quite a few of them to eventually be exposed to any objectionable ideas being identified by this experiment.

Thus, there is a kind of insidiousness to the infrequency of exposing such ideas in these series. The infrequency facilitates a subtle, continual reinforcement of certain assumptions in regular viewers, while casual, less frequent viewers may notice nothing at all.

There is also wide variability in how these ideas are presented. It took me a while to puzzle through the results and form a perspective on what I found. I did not end up finding what I thought I would find, which may actually be an indicator that this was a reasonably objective experiment. As I was auditing the results, I began to observe enough patterns that I was able to develop a kind of taxonomy to organize my thinking about the approaches TV shows take, when handling these sensitive ideas. There are multiple ways in which these subjects are introduced to children in TV shows.

• Blatant and absurd indoctrination - This is the most egregious approach and it is in-your-face in support of ideas that are damaging to children and subversive of the moral calculus of Christian parents. The most blatant offender in this area is Blue’s Clues. The video below provides one example. But Blue’s Clues openly and outspokenly celebrates Pride month and Pride parades. The video clip below shows a very young girl repeating LGBTQ cant about which, it is hard for me to believe, she has any actual understanding. There is something essentially abusive about using children to propagandize ideas regarding sexuality that they have no business occupying their minds with at this point in their lives.

• Assumptive and fleeting - This approach introduces possibly objectionable ideas by just assuming their normality, for the most part, and fleetingly including them in the overall story line. In the video below, I provide examples from SpongeBob SquarePants and Peppa the Pig. In SpongeBob, the idea of two men marrying each other and having a child is introduced, in passing, as part of describing a family tree. In Peppa, one of the characters mentions that she has two moms.

• Sexual Ambiguity - This approach involves, without comment, using characters that are sexually ambiguous. One example of this is the series My Little Pony. That series presents subtle and persistent head-nods to sexual ambiguity and homosexuality generally. The clip in the video below was identified by the AI model as an example of a dialog that suggested same-sex attraction. That’s because the AI computes the sex of one of the characters as being ‘male’. One of the characters is clearly female, but the sex of the other character is unclear. The AI model maintains an attention window for evaluating context that is 128,000 tokens wide. The means the model can compute context within a two-hour window of spoken human dialog. So the model identifying one of the characters as male wasn’t just based on this narrow scene alone. Anyone familiar with My Little Pony lore would know that the ambiguous character in the included clip is ostensibly female. But the sexual ambiguity is, I think, the point being made by the artists and script writers. The voice, appearance, and mannerisms of one of the characters is ambiguous in regard to sex. The reader can judge for himself after watching the video clip.

• Questionable Topics - The clip of Bluey which I included contains a humorous depiction of the father wearing a kind of pouch on his stomach, inside which the younger of his children is riding. The idea is for the father to experience what it’s like to be pregnant. The words “pregnant man” and “pregnant people” are used in the dialog. Increasingly, in the broader culture, this kind of language is a coded form of transgender affirmation. The scene is all very humorous and is not explicitly suggesting that the dad can really be pregnant. But, in our current moment of gender confusion, one can’t help but wonder whether this kind of dialog and storyline normalizes something that will inevitably be reinforced by more explicit and malevolent messaging from other sources. If this were 1970, any questions might be dismissed with “it’s just comedy”. But in 2024, it should at least raise questions in the minds of thoughtful viewers.

• Mirroring Cultural Assumptions - The episode of Bluey in this section was one that the AI flagged which, at first glance, I don’t think I would have considered. The episode deals humorously with the age-old competition between girls and boys, long depicted in a variety of ways by the media. But, in this particular case, it does reinforce the trendy view that boys and girls are interchangeable in terms of physical strength. This is culturally consistent with the entire girl-boss theming easily found in mass media, especially the super-hero movies where female characters are magically able to physically overpower rooms full of male attackers. Aside from the falsity of the view that male and female strength is equivalent, there are actually real-world consequences for engaging in such delusions. Just this week, a female boxer at the Olympics (don’t get me started on the unattractiveness of female boxing) had the snot beat out of her by a male opponent because the International Olympic Committee has embraced the fantasy that gender is merely a matter of personal preference. The necessary corollary to such thinking is that any physical strength differential between men and women are simply not considerations that should be taken into account. Again, in 1970 this clip would be innocuous and humorous. But now that men are beating up women in the Olympics, casually mirroring the underlying cultural assumptions in children’s TV seems like a catastrophically bad idea.

• Transcription Errors - Finally, I include in this video of clips, an interesting example of the AI being misled due to a transcription error made by the automated transcriber. The transcriber is also an AI model that can convert spoken words into text. The model transcribed the audio to say “she wants to marry mum”, when the character had actually said “you want to marry mum”. Because of the transcription error, the language model flagged this episode as being related to same-sex attraction. If you listen closely to the audio you can see why the transcription model made the mistake. This illustrates that all of the models involved in this exercise are error-prone. So, though technology like this can find some potential needles in the hay stack of children’s content, moral discernment still entails careful human examination of each possible “needle” on a individual basis.

Tentative Conclusions

So what have I concluded regarding the content of children’s TV, as well as the suitability of AI for characterizing that content?

It seems unavoidable that the conditioning going on with children’s programming is easing the way toward children eventually accepting radical claims regarding gender and human sexuality. That such conditioning is not constant is, on the one hand, a relief. But I wonder if the background persistence of it, as a dripping, ambient theme, is not more insidious even than the blatant absurdities of something like Blue’s Clues. Something, or some combination of things, has paved the way for millions of young people to casually embrace a set of assumptions about sex and gender that are, literally, unreal. Nothing about my efforts looking into this has disabused me of the suspicion that children’s TV is a very real contributor to that unreality.

Stepping back from the technological questions for a moment, it seems like common sense to observe that, as the broader culture is emptied out of Christian presuppositions, it is inevitable that children’s programming will predominately reflect the prevailing cultural assumptions, which will often be adversarial to Christian ideals. This implies that, as the culture becomes ever less Christian, Christian consumption of mainstream entertainment will entail marinating ourselves in ideas that are adversarial to Christian faith. I will note that I do make a distinction between passive consumption of media for entertainment, and active analysis of media for understanding our current moment.

This is not a call for Christians to abandon the arts. On the contrary, it is a call to be discerning and selective, embracing and promoting art that is true and beautiful while rejecting the ugliness and lies foisted upon us as entertainment. The Olympic opening ceremonies were the latest case-in-point. Christians need to return to the practice of our ancestors by becoming far more prolific in making art that is true to what is real - art that reinforces the beautiful and the good.

As far as the technology itself is concerned, it is currently useful but far from perfect. My own experiments found it to be very useful and productive for sorting through information at a volume that no individual could otherwise hope to do. But blindly trusting its results is unwise, especially as long as there is ambiguity surrounding how it computes its conclusions. The most recent models, with their very large attention windows, are just now becoming more suitable for this kind of analysis. But the very essence of these models is to compute their results based on statistics rather than actual meaning. To fully trust the models, one would have to accept that statistics are interchangeable with meaning. That way lies dragons.

In my own efforts, I was only able to use these models to whittle down the massive corpus of episodes to a smaller list for my own human consideration. Ultimately, though, it still boiled down to a time-consuming exercise in discernment by a flesh-and-blood human being. Though the language models were able to reduce some of that effort, they are simply not a replacement for actual human insight.

On a personal note, there is something tedious and mind-numbing about the vacuous plots one finds in contemporary children’s TV. Completely aside from the obvious questions about gender and sexuality, the absurd triviality of the plots and ideas suggests something about how the makers of these shows perceive their audience. In reality, children are intensely moral creatures, with heroic aspirations and an appetite for achievement. We would do well just now, in selecting our children’s entertainment, to consider the words of Russell Kirk.

“In a violent time, it is prudent to rear children on tales of peril — and of heroism. If enough of the rising generation take the heroes of fantasy for their exemplars, the wolf will find sustenance less readily. ‘What sharp teeth you have!’ ‘The better to eat you with my dear.’ Give us more woodcutters, in the nick of time.”